Claude Opus 4.5: The $5 AI Model That Outscores Human Engineers at Coding

Anthropic announced Claude Opus 4.5, described as its most intelligent model yet with particular strength in solving coding tasks. The release represents a significant strategic move in Anthropic's product lineup, completing a three-tier offering alongside the faster Sonnet 4.5 and more compact Haiku 4.5 models.

Coding Capabilities & Performance

Opus 4.5 achieves an 80.9% accuracy score on SWE-Bench Verified, surpassing competing models including OpenAI's GPT-5.1 and Google's newly released Gemini 3. Beyond benchmark scores, Anthropic tested the model against its internal engineering assessment designed for prospective performance engineering hires.

Using parallel test-time compute, Opus 4.5 scored higher than any human candidate on the exam, and without a time limit matched the performance of the best-ever human candidate when used within Claude Code.

The company also demonstrated token efficiency advantages. At medium effort level, Opus 4.5 matches Sonnet 4.5's best SWE-bench score while using 76% fewer output tokens, and at the highest effort level exceeds Sonnet performance by 4.3 points while still using 48% fewer tokens.

Pricing Revolution

API pricing for Opus 4.5 is set at $5 per million input tokens and $25 per million output tokens, down substantially from the previous Opus 4.1 rates of $15 and $75.

This represents a roughly 67% reduction that fundamentally changes Opus accessibility. The new pricing brings Anthropic closer to competitors while maintaining a premium position, with OpenAI's GPT-5.1 at $1.25/$10 and Google's Gemini 3 Pro at $2-$4 per million input tokens.

Developer Platform Enhancements

Anthropic introduced an effort parameter enabling developers to control computational allocation across thinking, tool calls, and responses to balance performance with latency and cost. The company also released compaction control mechanisms for managing context efficiently during extended agentic tasks.

Claude Code gains enhancements including an improved plan mode that builds more precise plans and asks clarifying questions upfront, and is now available in Anthropic's desktop application. Additionally, for Claude app users, long conversations no longer hit a wall through automatic context summarization, allowing chats to continue indefinitely.

Enterprise Capabilities & Early Feedback

GitHub's chief product officer reported early testing shows Opus 4.5 surpasses internal coding benchmarks while cutting token usage in half, and is especially suited for code migration and code refactoring.

Rakuten, testing the model on office task automation, found that their agents autonomously refined capabilities achieving peak performance in 4 iterations while other models couldn't match that quality after 10.

Multi-Model Integration & Ecosystem

Opus 4.5 is available in Microsoft Foundry, GitHub Copilot paid plans, and Microsoft Copilot Studio. The model is also accessible through major cloud platforms including Amazon Bedrock, Google Cloud Vertex AI, and Microsoft Azure, alongside Anthropic's direct API.

Safety & Security Features

Anthropic claims state-of-the-art prompt injection resistance, representing significant progress on a problem the industry has largely deprioritized through training alone. This addresses a critical concern for deploying agents with tool access in sensitive enterprise environments.

Market Context

The launch occurs within an intensely competitive timeframe. The release follows Google's Gemini 3 and OpenAI's GPT-5.1, demonstrating the rapid pace of competition in enterprise AI. The launch marks Anthropic's third major model release in two months, following Sonnet 4.5 in September and Haiku 4.5 in October